small spaces question for Jochen

[Jonathan S. Shapiro]

It has occurred to me that if PTE invalidation is handled with
sufficient care, pages in small address spaces can safely be marked
global.

The working assumption here is that small processes own their portions
of the address space for relatively long periods of time.  Given that
the segment registers provide appropriate bounds checking, the
corresponding portion of the address space really only needs to be
invalidated when there is a permissions change in the small space
(relatively rare) and/or when a small space range is reassigned
(i.e. when a new process is assigned to that range).

The advantages seem potentially significant for the kinds of things
that in monolithic kernels would be handled by dynamically loaded
kernel modules.

Have you investigated this possibility?


shap